Ldsd: Lemon Surface Defect Detection System Based on Decision Tree

Conventionally, fruit sorting was predominantly carried out manually by workers, who relied on visual inspection to detect and classify surface defects of the fruits, and used measuring tools to assess their diameter and weight. However, with the increasing labor costs and the manufacturers' pursuit of higher production efficiency, people have begun to adopt a method that integrates machine vision with sensors to perform fruit inspection tasks. Numerous researches related to machine learning have emerged around the world. It has been adapted to fields including disease prediction, market prediction, industrial procedures etc. The use of machine learning for surface defect detection in fruit sorting is still an area that requires further research. Among the existing researches, only limited number of models are used, resulting in the lack of comparison of accuracy. In this article, our team compares the accuracy of four models: Decision Tree, Support Vector Machine (SVM), Naïve Bayes, KNearest Neighbors (KNN). This paper will explain why these four models were selected as candidate models. After selecting the most suitable model (Decision Tree) for the task of lemon surface defect detection, this paper designs a corresponding vision system for capturing lemon images, develops a diameter detection algorithm for measuring the diameter of lemons, designs a set of sensors and their accompanying data acquisition card for obtaining lemon weight data, and refines the design of a high-quality lemon grading system for classifying the results. In the end, the model based on decision tree outperformed other models with an accuracy of 0.9520, an area under curve of 0.9416, a <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$F 1$</tex> score of 0.9484, a precision of 0.9482 and a recall of 0.9520. The system will be of high efficiency and significantly reduce the cost of labor cost in order to cope with increasingly high market demand and assist manufacturers stand out in the fierce competition of modern market.